Publish Agent Failure Atlas Static Space v0.2.0
Browse files
README.md
CHANGED
|
@@ -1,10 +1,54 @@
|
|
| 1 |
---
|
| 2 |
title: Agent Failure Atlas
|
| 3 |
-
emoji:
|
| 4 |
-
colorFrom:
|
| 5 |
colorTo: indigo
|
| 6 |
sdk: static
|
|
|
|
|
|
|
| 7 |
pinned: false
|
|
|
|
|
|
|
| 8 |
---
|
| 9 |
|
| 10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
title: Agent Failure Atlas
|
| 3 |
+
emoji: 🧭
|
| 4 |
+
colorFrom: yellow
|
| 5 |
colorTo: indigo
|
| 6 |
sdk: static
|
| 7 |
+
app_file: index.html
|
| 8 |
+
fullWidth: true
|
| 9 |
pinned: false
|
| 10 |
+
license: apache-2.0
|
| 11 |
+
short_description: Private, client-side analysis of AI-agent trace failures.
|
| 12 |
---
|
| 13 |
|
| 14 |
+
# Agent Failure Atlas — Static Space
|
| 15 |
+
|
| 16 |
+
A browser-only showcase for the Open Agent Failure Atlas.
|
| 17 |
+
|
| 18 |
+
- Trace files are parsed and analyzed locally in the browser.
|
| 19 |
+
- The Space makes no model calls and does not upload user-provided traces.
|
| 20 |
+
- It supports Hugging Face Session Trace Simple Format (STS), one-row normalized datasets, and OpenAI-style message JSONL.
|
| 21 |
+
- The complete Python CLI, native adapters, release tooling, and tests remain in the main repository.
|
| 22 |
+
|
| 23 |
+
## Public artifacts
|
| 24 |
+
|
| 25 |
+
- [GitHub](https://github.com/Solasticeaistudio/agent-failure-atlas)
|
| 26 |
+
- [Benchmark Dataset](https://huggingface.co/datasets/solsticestudioai/agent-failure-atlas-benchmark)
|
| 27 |
+
|
| 28 |
+
## Deploy
|
| 29 |
+
|
| 30 |
+
Create a Hugging Face Space using the **Static** SDK, then upload the contents of this directory to the root of the Space repository. The YAML front matter above configures `sdk: static` and `app_file: index.html`.
|
| 31 |
+
|
| 32 |
+
No build step or server dependency is required.
|
| 33 |
+
|
| 34 |
+
## Local verification
|
| 35 |
+
|
| 36 |
+
```bash
|
| 37 |
+
python -m http.server 8000
|
| 38 |
+
# Open http://localhost:8000
|
| 39 |
+
|
| 40 |
+
node --test tests/core.test.mjs
|
| 41 |
+
```
|
| 42 |
+
|
| 43 |
+
## Scope
|
| 44 |
+
|
| 45 |
+
The browser implementation mirrors the six deterministic detector families used by the Python MVP:
|
| 46 |
+
|
| 47 |
+
- scope and unauthorized-tool violations
|
| 48 |
+
- approval bypass
|
| 49 |
+
- prompt injection in tool output
|
| 50 |
+
- runaway repeated actions
|
| 51 |
+
- silent or unhandled tool failure
|
| 52 |
+
- secret exposure
|
| 53 |
+
|
| 54 |
+
The bundled 20-session result is a synthetic software smoke benchmark. A perfect fixture score does not establish real-world detector accuracy or model safety.
|